Operation and maintenance processing method and device, electronic equipment and storage medium

Abstract

The invention provides an operation and maintenance processing method and device, electronic equipment and a storage medium, and relates to the field of intelligent operation and maintenance. The method comprises the following steps: firstly, classifying system data of a target system into high-sensitivity data and low-sensitivity data, storing the high-sensitivity data into a local storage pool, and storing the low-sensitivity data into a cloud storage pool; then, when an operation and maintenance request sent by a user side is received, the operation and maintenance request can be analyzed to obtain request metadata, task sensitivity detection is performed on the request metadata, and a task type corresponding to the operation and maintenance request is determined. If the task type is a high-sensitivity task, processing the operation and maintenance request by using a local pre-training model and high-sensitivity data; and if the task type is a low-sensitivity task, processing the operation and maintenance request by using a cloud pre-training model and low-sensitivity data. Therefore, the sensitive data of the target system can be prevented from clouding, so that the operation and maintenance security can be improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance, and in particular to an operation and maintenance processing method, apparatus, electronic device and storage medium. Background Technology

[0002] With the continuous development of artificial intelligence (AI) technology, many enterprises are introducing AI models for intelligent operation and maintenance of their systems. To meet computing power requirements, these AI models are typically deployed in the cloud. However, this requires uploading sensitive data from the enterprise system to the cloud, posing significant privacy concerns. Summary of the Invention

[0003] The purpose of this invention is to provide an operation and maintenance processing method, device, electronic device and storage medium that can avoid uploading highly sensitive data of the target system to the cloud during the operation and maintenance process, thereby improving the security of intelligent operation and maintenance.

[0004] To solve the above-mentioned technical problems, the present invention provides an operation and maintenance processing method, including:

[0005] The system data of the target system is acquired, classified into highly sensitive data and low sensitive data, and the highly sensitive data is stored in a local storage pool and the low sensitive data is stored in a cloud storage pool.

[0006] When a maintenance request is received from a user, the maintenance request is parsed to obtain request metadata, and task sensitivity detection is performed on the request metadata to determine the task type corresponding to the maintenance request.

[0007] If the task type is a highly sensitive task, the local pre-trained model is used to generate an operation and maintenance dialogue result based on the operation and maintenance request and the highly sensitive data in the local storage pool and send it to the user terminal, or the local pre-trained model is used to perform operation and maintenance operations on the target system based on the operation and maintenance request and the highly sensitive data; wherein, the target system, the local storage pool, and the local pre-trained model are all located in the same security domain;

[0008] If the task type is a low-sensitivity task, the cloud-pre-trained model is used to generate an operation and maintenance dialogue result based on the operation and maintenance request and the low-sensitivity data in the cloud storage pool, and then the result is sent to the user terminal.

[0009] Optionally, the operation and maintenance request is parsed to obtain request metadata, including:

[0010] The operation and maintenance request is subjected to intent recognition to obtain user intent metadata;

[0011] The operation and maintenance request is parsed to extract sensitive fields and obtain sensitive metadata.

[0012] Query the client information and request behavior information of the user terminal corresponding to the maintenance request, and use the client information and request behavior information as network context metadata;

[0013] The user intent metadata, the data-sensitive metadata, and the network context metadata are integrated into the request metadata.

[0014] Optionally, task sensitivity detection is performed on the request metadata to determine the task type corresponding to the operation and maintenance request, including:

[0015] The semantic similarity of the user intent metadata with the preset intent information is used to determine the target intent information that best matches the user intent metadata, and the matching degree of the target intent information is used as the intent matching degree.

[0016] The field sensitivity is obtained by calculating the ratio of the number of sensitive fields to the total number of fields in the operation and maintenance request.

[0017] The network context metadata is used to perform behavior pattern matching with preset behavior information to determine the preset behavior information that best matches the network context metadata, and the matching degree of the preset behavior information is used as the behavior pattern matching degree.

[0018] Using the preset weights corresponding to the intent matching degree, the field sensitivity, and the behavior pattern matching degree, the intent matching degree, field sensitivity, and behavior pattern matching degree are weighted and fused to obtain a sensitivity value;

[0019] If the sensitivity value is less than the first preset threshold, then the task type is determined to be a low-sensitivity task;

[0020] If the sensitivity value is not less than the first preset threshold, then the task type is determined to be a highly sensitive task.

[0021] Optionally, it also includes:

[0022] When input adjustment information is received, the preset behavior information, the sensitive field to be extracted, the preset behavior information, or the preset weight are adjusted using the adjustment information.

[0023] Optionally, before generating maintenance dialogue results based on the maintenance request and highly sensitive data in the local storage pool using a locally pre-trained model and sending them to the user terminal, or before performing maintenance operations on the target system using the locally pre-trained model based on the maintenance request and the highly sensitive data, the method further includes:

[0024] A pre-trained risk detection model is used to detect risks in the request metadata to determine the risk value corresponding to the operation and maintenance request.

[0025] If the risk value is less than the second preset threshold, then proceed to the step of generating an operation and maintenance dialogue result based on the operation and maintenance request and the highly sensitive data in the local storage pool using the local pre-trained model and sending it to the user terminal, or performing operation and maintenance operations on the target system based on the operation and maintenance request and the highly sensitive data using the local pre-trained model.

[0026] If the risk value is not less than the second preset threshold, the maintenance request will be intercepted.

[0027] Optionally, it also includes:

[0028] Training data is generated using the requested metadata and manually labeled information;

[0029] The pre-trained risk detection model is incrementally trained using the training data.

[0030] Optionally, the operation and maintenance request is parsed to obtain request metadata, and task sensitivity detection is performed on the request metadata to determine the task type corresponding to the operation and maintenance request, including:

[0031] The operation and maintenance request is decomposed into tasks to obtain at least two sub-tasks that need to be executed to process the operation and maintenance request, and the task metadata of the sub-tasks is extracted.

[0032] The task metadata is subjected to task sensitivity detection to determine the task type corresponding to each subtask, so that the subtask is scheduled to be processed by the local pre-trained model or the cloud pre-trained model according to the task type.

[0033] Optionally, storing the low-sensitivity data in a cloud storage pool includes:

[0034] Sensitive fields in the low-sensitivity data are de-identified;

[0035] The desensitized, low-sensitivity data is saved to the cloud storage pool.

[0036] Optionally, it also includes:

[0037] The first training data is set using the highly sensitive data in the local storage pool, and the local pre-trained model is fine-tuned using the first training data.

[0038] The second training data is set using low-sensitivity data in the cloud storage pool. The cloud pre-trained model is fine-tuned using the second training data. During the training process of the cloud pre-trained model, noise is added to the training gradient generated by the cloud pre-trained model, and the parameters of the cloud pre-trained model are updated based on the training gradient with added noise.

[0039] The present invention also provides an operation and maintenance processing device, comprising:

[0040] A classification storage module is used to acquire system data of the target system, classify the system data into high-sensitivity data and low-sensitivity data, and store the high-sensitivity data in a local storage pool and the low-sensitivity data in a cloud storage pool.

[0041] The request parsing module is used to parse the operation and maintenance request sent by the user to obtain request metadata, and to perform task sensitivity detection on the request metadata to determine the task type corresponding to the operation and maintenance request.

[0042] The high-sensitivity task processing module is used to generate an operation and maintenance dialogue result based on the operation and maintenance request and the high-sensitivity data in the local storage pool and send it to the user terminal if the task type is a high-sensitivity task, or to perform operation and maintenance operations on the target system based on the operation and maintenance request and the high-sensitivity data using the local pre-trained model.

[0043] The low-sensitivity task processing module is used to generate an operation and maintenance dialogue result based on the operation and maintenance request and the low-sensitivity data in the cloud storage pool, and send it to the user terminal if the task type is a low-sensitivity task.

[0044] The present invention also provides an electronic device, comprising:

[0045] Memory, used to store computer programs;

[0046] A processor is used to implement the operation and maintenance processing method described above when executing the computer program.

[0047] The present invention also provides a computer-readable storage medium storing computer-executable instructions, which, when loaded and executed by a processor, implement the operation and maintenance processing method described above.

[0048] This invention provides an operation and maintenance (O&M) processing method, comprising: acquiring system data of a target system, classifying the system data into highly sensitive data and low-sensitive data, storing the highly sensitive data in a local storage pool, and storing the low-sensitive data in a cloud storage pool; when receiving an O&M request sent by a user terminal, parsing the O&M request to obtain request metadata, and performing task sensitivity detection on the request metadata to determine the task type corresponding to the O&M request; if the task type is a highly sensitive task, generating an O&M dialogue result based on the O&M request and the highly sensitive data in the local storage pool using a local pre-trained model and sending it to the user terminal, or performing O&M operations on the target system based on the O&M request and the highly sensitive data using the local pre-trained model; wherein the target system, the local storage pool, and the local pre-trained model are all located in the same security domain; if the task type is a low-sensitive task, generating an O&M dialogue result based on the O&M request and the low-sensitive data in the cloud storage pool using a cloud pre-trained model and sending it to the user terminal.

[0049] The beneficial effects of this invention are as follows: First, this invention can classify the system data of the target system into highly sensitive data and low-sensitivity data, storing highly sensitive data in a local storage pool and low-sensitivity data in a cloud storage pool, thus achieving differentiated storage of highly sensitive and low-sensitivity data. Subsequently, when a maintenance request is received from a user, this invention can parse the maintenance request to obtain request metadata, and perform task sensitivity detection on the request metadata to determine the task type corresponding to the maintenance request. If the task type is a highly sensitive task, a local pre-trained model can be used to generate maintenance dialogue results based on the maintenance request and the highly sensitive data in the local storage pool and send them to the user, or the local pre-trained model can be used to perform maintenance operations on the target system based on the maintenance request and the highly sensitive data; that is, the local pre-trained model can be used to access the highly sensitive data of the target system and perform more sensitive maintenance processing on the target system. If the task type is a low-sensitivity task, a cloud pre-trained model can be used to generate maintenance dialogue results based on the maintenance request and the low-sensitivity data in the cloud storage pool and send them to the user. In this way, by distinguishing and storing system data of different sensitivities in the target system, and by using local pre-trained models and cloud pre-trained models to perform operation and maintenance tasks of different sensitivities on the target system, the present invention can avoid sensitive data of the target system being uploaded to the cloud. Furthermore, the present invention can set the target system, local storage pool, and local pre-trained model in the same security domain, ensuring that the target system and its highly sensitive data are effectively isolated, thereby improving the security of intelligent operation and maintenance.

[0050] The present invention also provides an operation and maintenance processing device, an electronic device, and a computer-readable storage medium, which have the above-mentioned beneficial effects. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0052] Figure 1 A flowchart of an operation and maintenance processing method provided in an embodiment of the present invention;

[0053] Figure 2 A structural block diagram of an intelligent operation and maintenance system provided in an embodiment of the present invention;

[0054] Figure 3 A schematic diagram of an operation and maintenance processing framework provided in an embodiment of the present invention;

[0055] Figure 4 This is a schematic diagram illustrating local deployment and cloud deployment as provided in an embodiment of the present invention;

[0056] Figure 5 A timing diagram of an operation and maintenance processing method provided in an embodiment of the present invention;

[0057] Figure 6 This is a structural block diagram of an operation and maintenance processing device provided in an embodiment of the present invention;

[0058] Figure 7 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] With the continuous development of artificial intelligence (AI) technology, many enterprises are introducing AI models for intelligent operation and maintenance of their systems. To meet computing power requirements, these AI models are typically deployed in the cloud. However, this requires uploading sensitive data from the enterprise system to the cloud, posing significant privacy concerns.

[0061] In view of this, in order to improve the security of intelligent operation and maintenance and avoid sensitive system data from being uploaded to the cloud, the present invention can provide an operation and maintenance processing method. By classifying user operation and maintenance requests, and using a local pre-trained model to process highly sensitive requests and a cloud pre-trained model to process low-sensitivity requests, the local pre-trained model and the cloud pre-trained model can be used to distinguish and process requests of different sensitivity types, thereby avoiding highly sensitive system data from being uploaded to the cloud and improving the security of intelligent operation and maintenance.

[0062] For easier understanding, please refer to Figure 1 , Figure 1 This is a flowchart illustrating an operation and maintenance processing method provided in an embodiment of the present invention. The executing entity of this method can be an intelligent operation and maintenance system. This method may include:

[0063] S10. Obtain system data from the target system, classify the system data into highly sensitive data and low-sensitivity data, and store the highly sensitive data in the local storage pool and the low-sensitivity data in the cloud storage pool.

[0064] In this embodiment, the target system is the system that requires intelligent operation and maintenance (O&M). The target system can be any type of system, such as a business system or a network system. The system data of the target system is the data required for performing intelligent O&M, which allows the intelligent O&M system to understand the operating status of the target system. This embodiment does not limit the specific content of the system data; for example, it may include the operating data of each device in the system, telemetry data (such as network device status, performance indicators, and traffic information), operation logs, configuration files, fault records, etc.

[0065] It is important to note that the system data mentioned above includes both highly sensitive and low-sensitivity data. High-sensitivity data is typically the core data of the target system; its leakage would not only severely compromise the operational security of the target system but also potentially seriously jeopardize the information security and interests of the enterprise. Low-sensitivity data refers to non-core data of the target system; its leakage will not affect the operational security of the target system or will only have a minor impact on the information security of the enterprise. For example, in network systems, confidential information such as operational data, telemetry data, operational logs, and configuration files are considered highly sensitive data, while publicly available information such as network equipment maintenance manuals are considered low-sensitivity data.

[0066] Understandably, to prevent the leakage of core confidential information of the target system, the aforementioned highly sensitive information should not be uploaded to the cloud but should be properly stored within the enterprise. Therefore, this embodiment can set up a local storage pool on the enterprise's premises and a cloud storage pool in the cloud. When processing system data, the data can be classified into highly sensitive data and low-sensitivity data. High-sensitivity data is stored in the local storage pool, while low-sensitivity data is stored in the cloud storage pool, thus differentiating the storage of system data according to its sensitivity. That is, highly sensitive data will not be uploaded to the cloud storage pool through an external network, thereby ensuring the security of highly sensitive data.

[0067] It should be noted that this embodiment does not limit how the system data is classified. For example, since the type of system data is determined, classification rules can be manually set, and the system data can be classified into high-sensitivity data and low-sensitivity data according to these rules. Alternatively, a classification model can be trained and deployed, and used to classify the system data into high-sensitivity data and low-sensitivity data. This classification model can be a machine learning model, such as a support vector machine, decision tree model, or BERT model.

[0068] To further enhance information security, this embodiment can also perform anonymization processing on sensitive fields within low-sensitivity data when storing low-sensitivity data in a cloud storage pool. Specifically, sensitive fields (such as IP addresses, passwords, and user accounts) can be automatically identified and anonymized using placeholders or hash encoding. Subsequently, the anonymized low-sensitivity data is saved to the cloud storage pool.

[0069] Based on this, storing low-sensitivity data in a cloud storage pool can include:

[0070] S11: Desensitize sensitive fields in low-sensitivity data;

[0071] S12: Save the desensitized low-sensitivity data to the cloud storage pool.

[0072] It should be noted that this embodiment does not limit the specific sensitive fields, and can be set according to actual application needs. For example, a sensitive field dictionary can be preset, which contains all sensitive fields, and this dictionary can be used to identify sensitive fields.

[0073] Furthermore, the aforementioned system data can be collected from multiple data sources within the target system, using different collection methods. This embodiment does not limit the specific collection method; for example, it can be SNMP, TR-181 / TR-369, NetFlow, Syslog, etc. After data collection is completed, the obtained raw system data can be cleaned, such as performing data standardization operations (unifying timestamps, field naming, format conversion, etc.) and data tagging operations (mapping status + logs + user queries to configuration or fault handling actions) to ensure that the system data content is clear and the data format is standardized.

[0074] Furthermore, in this embodiment, a data collector can be set up in the intelligent operation and maintenance system to collect system data from the target system, clean, standardize and classify the system data, and save the obtained highly sensitive data and low sensitive data to the local storage pool and the cloud storage pool, respectively.

[0075] S20. When a maintenance request is received from the user, the maintenance request is parsed to obtain the request metadata, and the task sensitivity detection is performed on the request metadata to determine the task type corresponding to the maintenance request.

[0076] In this embodiment, the user terminal can send an operation and maintenance request to the intelligent operation and maintenance system, which can indicate the user's operation and maintenance intention. It is worth noting that the user can describe the operation and maintenance request using natural language. For example, for a network system, the user can send "What is the recent operating status of a certain network device?" or "Adjust a certain network configuration of a certain network device", etc. This embodiment does not limit the specific content and form of the operation and maintenance request, and can be set according to actual application needs.

[0077] It is worth noting that, when processing different maintenance requests, the intelligent maintenance system may use system data of varying sensitivity and may perform operations on the target system of varying sensitivity levels. Highly sensitive system data should not be accessed by the cloud, and highly sensitive system operations should not be executed by the cloud. Therefore, this embodiment needs to distinguish between high-sensitivity and low-sensitivity tasks, processing high-sensitivity tasks locally and low-sensitivity tasks in the cloud. Thus, this embodiment needs to parse the maintenance request to obtain request metadata in this step, perform task sensitivity detection on the request metadata, determine the task type corresponding to the maintenance request, and then schedule the maintenance request to be processed locally or in the cloud. This embodiment can set up a dialogue interface in the intelligent maintenance system to determine the task type corresponding to the maintenance request and schedule it to be processed locally or in the cloud.

[0078] Specifically, since users can describe their maintenance requests using natural language, and natural language requests are not conducive to detection, this embodiment first extracts key information from the maintenance request and standardizes it into request metadata, then uses the request metadata for task sensitivity detection. This embodiment does not limit the specific type of request metadata and can set it according to actual application needs. For example, in one implementation, the request metadata may include user intent metadata, data-sensitive metadata, and network context metadata. User intent metadata is used to indicate the user's maintenance intent, such as querying information or modifying configurations. Data-sensitive metadata is used to indicate sensitive fields in the maintenance request, such as IP address, username, and password. Network context metadata may contain user client information and request behavior information. User client information may include the user's name, ID, and permission level, while request behavior information may include the user's recent historical request content and request frequency. By setting the above content, this embodiment can classify user maintenance requests from aspects such as user intent, sensitive fields, user information, and user permissions.

[0079] Based on this, parsing the operation and maintenance request yields request metadata, which may include:

[0080] S211: Perform intent recognition on maintenance requests to obtain user intent metadata;

[0081] S212: Perform field parsing on the operation and maintenance request, extract sensitive fields from the operation and maintenance request, and obtain data-sensitive metadata;

[0082] S213: Query the client information and request behavior information of the user terminal corresponding to the operation and maintenance request, and use the client information and request behavior information as network context metadata;

[0083] S214: Integrate user intent metadata, data-sensitive metadata, and network context metadata into request metadata.

[0084] Furthermore, this embodiment can perform task sensitivity detection on the request metadata to determine the task type of the operation and maintenance request. This embodiment does not limit the specific method of task sensitivity detection. For example, a pre-trained classification model can be used to classify the request metadata to determine the task type. The pre-trained classification model is a machine learning model that has been pre-trained and can classify request metadata. For another example, to perform task sensitivity detection from multiple directions and improve the adjustability of task sensitivity detection, this embodiment can separately detect user intent metadata, data sensitive metadata, and network context metadata to obtain intent matching degree, field sensitivity, and behavior pattern matching degree. The intent matching degree, field sensitivity, and behavior pattern matching degree are then weighted and summed to obtain a sensitivity value. It is then determined whether the sensitivity value is less than a preset threshold. If it is less, the task type is determined to be a low-sensitivity task; if it is not less, the task type is determined to be a high-sensitivity task.

[0085] Specifically, for intent matching, preset intent information corresponding to highly sensitive task types can be set, and user intent metadata can be matched with the preset intent information to determine the maximum intent matching degree corresponding to the user intent metadata. This maximum intent matching degree is then used as the aforementioned intent matching degree. For field sensitivity, the ratio of the number of sensitive fields to the total number of fields in the maintenance request is calculated to obtain the field sensitivity. For behavior pattern matching, preset behavior information corresponding to highly sensitive task types (such as operation combinations formed by request behavior information, operation methods under specific user permissions, etc.) can be set, and network context metadata can be matched with the preset behavior information to determine the maximum behavior pattern matching degree corresponding to the network context metadata. This maximum behavior pattern matching degree is then used as the aforementioned behavior pattern matching degree. Thus, this embodiment can perform task sensitivity detection from three aspects: user intent, sensitive fields, and user behavior. Furthermore, the preset intent information, sensitive fields, preset behavior information, and weight values ​​can all be dynamically adjusted according to application requirements, improving the adjustability of task sensitivity detection.

[0086] Based on this, task sensitivity detection is performed on the request metadata to determine the task type corresponding to the operation and maintenance request, which may include:

[0087] S221: Perform semantic similarity matching between user intent metadata and preset intent information to determine the target intent information that best matches the user intent metadata, and use the matching degree of the target intent information as the intent matching degree;

[0088] S222: Calculate the ratio of the number of sensitive fields to the total number of fields in the operation and maintenance requests to obtain the field sensitivity;

[0089] S223: Use network context metadata and preset behavior information to perform behavior pattern matching, determine the preset behavior information that best matches the network context metadata, and use the matching degree of the preset behavior information as the behavior pattern matching degree.

[0090] S224: Using the preset weights corresponding to intent matching degree, field sensitivity and behavior pattern matching degree respectively, the intent matching degree, field sensitivity and behavior pattern matching degree are weighted and fused to obtain the sensitivity value;

[0091] S225: If the sensitivity value is less than the first preset threshold, the task type is determined to be a low-sensitivity task;

[0092] S226: If the sensitivity value is not less than the first preset threshold, then the task type is determined to be a high-sensitivity task.

[0093] It should be noted that this embodiment does not limit the specific value of the first preset threshold, and it can be set according to actual application needs.

[0094] Furthermore, the aforementioned preset behavioral information, sensitive fields to be extracted, preset behavioral information, or preset weights can all be adjusted as needed. Therefore, upon receiving input adjustment information, the preset behavioral information, sensitive fields to be extracted, preset behavioral information, or preset weights can be adjusted using the adjustment information, so that the detection content of task sensitivity detection can be adjusted according to actual needs.

[0095] Based on this, the method may also include:

[0096] S231: When receiving input adjustment information, adjust the preset behavior information, the sensitive field to be extracted, the preset behavior information or the preset weight using the adjustment information.

[0097] S30. If the task type is a highly sensitive task, the local pre-trained model is used to generate an operation and maintenance dialogue result based on the operation and maintenance request and the highly sensitive data in the local storage pool, and then sent to the user terminal; or the local pre-trained model is used to perform operation and maintenance operations on the target system based on the operation and maintenance request and the highly sensitive data. The target system, the local storage pool, and the local pre-trained model are all located in the same security domain.

[0098] In this embodiment, if the task type is determined to be a highly sensitive task, to avoid uploading highly sensitive data to the cloud and to prevent the cloud from performing sensitive operations on the target system, this embodiment can set up a local pre-trained model on-premises within the enterprise, and the highly sensitive task can be processed by the local pre-trained model. The local pre-trained model is a machine learning model pre-trained based on highly sensitive data for processing maintenance requests. For example, the local pre-trained model can be a language processing model (such as LLM, Large Language Model), a multimodal processing model, an AI agent, etc. When the dialogue interface determines that the maintenance request corresponds to a highly sensitive task, it can schedule the maintenance request to be processed by the local pre-trained model.

[0099] For highly sensitive tasks, this embodiment provides two processing methods: 1. Utilize a local pre-trained model to generate maintenance dialogue results based on maintenance requests and highly sensitive data in the local storage pool, and send the maintenance dialogue results to the user terminal. This processing method can effectively handle the user terminal's query and data analysis needs. 2. Utilize a local pre-trained model to perform maintenance operations on the target system based on maintenance requests and highly sensitive data. For example, the local pre-trained model can generate updated configuration information and send it to the target system, or it can generate operation scripts and execute them on the target system to perform operations. The target system can provide an automated script interface to support end-to-end maintenance closed loop. This processing method can effectively handle the user terminal's operational needs for the target system. Therefore, the local pre-trained model can handle highly sensitive tasks such as configuration generation, fault diagnosis, and sensitive data inference. Since this invention only allows the local pre-trained model to access highly sensitive data in the local storage pool and process highly sensitive tasks, and the local pre-trained model is set up locally in the enterprise, it can effectively prevent highly sensitive data from being uploaded to the cloud and sensitive operations from being performed on the target system by the cloud, thereby effectively improving the security of intelligent operation and maintenance and the operational security of the target system.

[0100] Furthermore, since the local pre-trained model can access highly sensitive data and perform highly sensitive operations, user access to the local pre-trained model should be effectively controlled to prevent malicious users from accessing it. To this end, when it is determined that a user-initiated maintenance request is a highly sensitive task, this embodiment can also perform risk detection on the user's request metadata to determine the risk value of the maintenance request. The risk value represents the degree of risk of the maintenance request; the higher the risk value, the higher the degree of risk. If the risk value is less than a preset threshold, this embodiment can determine that the maintenance request has a low risk and can be allowed to proceed and processed by the local pre-trained model. If the risk value is not less than the preset threshold, this embodiment can determine that the maintenance request has a high risk and needs to be intercepted. In this way, this embodiment can effectively intercept malicious highly sensitive tasks initiated by the user, thereby further improving the security of intelligent maintenance.

[0101] It should be noted that this embodiment does not limit the specific method of risk detection. For example, it can use rule matching to match the request metadata with preset rules and use the matching degree as the risk value. Alternatively, a pre-trained risk detection model can be used to detect the request metadata and generate a risk value. Here, the pre-trained risk detection model refers to a machine learning model that has been trained to detect whether the request metadata belongs to a high-risk type; for example, it can be a classification model, a BERT model, etc. In this embodiment, the predicted probability output by the pre-trained risk detection model to determine whether the request metadata belongs to a high-risk type can be used as the risk value. In this embodiment, a pre-trained risk detection model can be used for risk detection to improve the detection effect.

[0102] Based on this, before generating maintenance dialogue results using a locally pre-trained model based on maintenance requests and highly sensitive data in the local storage pool and sending them to the user terminal, or before performing maintenance operations on the target system using a locally pre-trained model based on maintenance requests and highly sensitive data, the following may also be included:

[0103] S311: Use a pre-trained risk detection model to perform risk detection on request metadata and determine the risk value corresponding to the operation and maintenance request;

[0104] S312: If the risk value is less than the second preset threshold, then proceed to the step of generating operation and maintenance dialogue results based on the operation and maintenance request and the highly sensitive data in the local storage pool using the local pre-trained model and sending them to the user terminal, or performing operation and maintenance operations on the target system based on the operation and maintenance request and the highly sensitive data using the local pre-trained model.

[0105] S313: If the risk value is not less than the second preset threshold, the operation and maintenance request will be intercepted.

[0106] Thus, this embodiment can use a local pre-trained model to process maintenance requests when the task type is determined to be a highly sensitive task and the user's maintenance request is low-risk, thereby effectively ensuring the security of intelligent maintenance.

[0107] It should be noted that risk detection for maintenance requests can also be performed via the dialog interface.

[0108] Furthermore, to ensure the detection performance of the pre-trained risk detection model, this embodiment can dynamically collect request metadata and its corresponding manually labeled information, and use the request metadata and manually labeled information to generate training data. The manually labeled information refers to manually reviewed labeling information indicating whether the request metadata corresponds to a high-risk type. Therefore, this embodiment can use the training data to incrementally train the pre-trained risk detection model to maintain its performance.

[0109] Based on this, the method may also include:

[0110] S321: Generate training data using request metadata and manually labeled information;

[0111] S322: Incrementally train the pre-trained risk detection model using training data.

[0112] Furthermore, to enhance the security of the intelligent operation and maintenance system, the target system, local storage pool, and local pre-trained model can reside within the same security domain. A security domain is a logically or physically isolated unit in network security, defined based on shared security requirements, used to implement unified protection strategies and control cross-domain data flows. This effectively isolates the target system, local storage pool, and local pre-trained model from the external environment, thereby improving security.

[0113] Furthermore, since the data collector can gather highly sensitive data and the dialogue interface needs to perform risk detection on the user end, meaning the data collector and dialogue interface are highly sensitive, they also need to be effectively protected. Therefore, the data collector and dialogue interface can also be placed in the same security domain as the target system, local storage pool, and local pre-trained model.

[0114] S40. If the task type is a low-sensitivity task, the cloud-pre-trained model is used to generate the operation and maintenance dialogue result based on the operation and maintenance request and the low-sensitivity data in the cloud storage pool and send it to the user terminal.

[0115] In this embodiment, if the task type is determined to be a low-sensitivity task, to avoid wasting local computing power, a cloud-based pre-trained model can be set up in the cloud, and the low-sensitivity task can be processed by the cloud-based pre-trained model. The cloud-based pre-trained model is a machine learning model pre-trained based on low-sensitivity data for processing operation and maintenance requests. For example, the local pre-trained model can be a language processing model (such as LLM, Large Language Model), a multimodal processing model, an AI agent, etc. When the dialogue interface determines that the operation and maintenance request corresponds to a low-sensitivity task, it can schedule the operation and maintenance request to the cloud-based pre-trained model for processing. For low-sensitivity tasks, the cloud-based pre-trained model generates operation and maintenance dialogue results based on the operation and maintenance request and the low-sensitivity data in the cloud storage pool, and sends the operation and maintenance dialogue results to the user terminal. This processing method can effectively handle the user's query and data analysis needs. Therefore, the cloud-based pre-processing model can handle low-sensitivity tasks such as statistical analysis, trend prediction, and cross-enterprise model inference. Since the cloud-based preprocessing model is only used to handle low-sensitivity queries and data analysis needs on the user side, it cannot access the high-sensitivity data of the target system or perform high-sensitivity operations on the target system. Therefore, it can improve the security of intelligent operation and maintenance, while avoiding the waste of the computing power that the enterprise has set up for local pre-trained models.

[0116] It should be noted that the cloud-based pre-trained model does not have operational permissions for the target system and cannot access the local storage pool. Therefore, this embodiment can effectively isolate the cloud-based pre-trained model from the highly sensitive data of the target system, thereby improving the security of the intelligent operation and maintenance system.

[0117] Furthermore, to ensure cloud security, cloud-based pre-trained models and cloud storage pools can also be set up within the same security domain. It's important to note that the security domain for cloud-based pre-trained models and cloud storage pools is different from the local security domain. For clarity, please refer to [link to relevant documentation]. Figure 2 , Figure 2 This is a structural block diagram of an intelligent operation and maintenance system provided in an embodiment of the present invention. It can be seen that the target system, data collector, local storage pool, local pre-trained model, and dialogue interface can be set in a first security domain on the enterprise's local premises, while the cloud storage pool and cloud pre-trained model can be set in a second security domain in the cloud. When a user sends a request to the dialogue interface, the dialogue interface needs to classify the request. If the request is a low-sensitivity request, the dialogue interface will schedule it for processing by the cloud pre-trained model in the second security domain. If the request is a high-sensitivity request, the dialogue interface will perform risk detection and authentication for the high-sensitivity request, and only after the risk detection and authentication are passed will the high-sensitivity request be scheduled for processing by the local pre-trained model in the first security domain, thereby ensuring the security of the intelligent operation and maintenance system.

[0118] Based on the above embodiments, the present invention first classifies the system data of the target system into highly sensitive data and low-sensitivity data, storing the highly sensitive data in a local storage pool and the low-sensitivity data in a cloud storage pool, thus achieving differentiated storage of highly sensitive and low-sensitivity data. Subsequently, when a maintenance request is received from a user, the present invention can parse the maintenance request to obtain request metadata, and perform task sensitivity detection on the request metadata to determine the task type corresponding to the maintenance request. If the task type is a highly sensitive task, a local pre-trained model can be used to generate maintenance dialogue results based on the maintenance request and the highly sensitive data in the local storage pool and send them to the user, or the local pre-trained model can be used to perform maintenance operations on the target system based on the maintenance request and the highly sensitive data; that is, the local pre-trained model can be used to access the highly sensitive data of the target system and perform more sensitive maintenance processing on the target system. If the task type is a low-sensitivity task, a cloud pre-trained model can be used to generate maintenance dialogue results based on the maintenance request and the low-sensitivity data in the cloud storage pool and send them to the user. In this way, by distinguishing and storing system data of different sensitivities in the target system, and by using local pre-trained models and cloud pre-trained models to perform operation and maintenance tasks of different sensitivities on the target system, the present invention can avoid sensitive data of the target system being uploaded to the cloud, thereby improving the security of intelligent operation and maintenance.

[0119] Based on the above embodiments, since user-submitted maintenance requests may be complex, directly determining the sensitivity of the maintenance request and scheduling local and cloud-based pre-trained models accordingly cannot achieve fine-grained control and scheduling, and may still lead to data leakage or wasted computing power. Therefore, this embodiment can also first decompose the maintenance request into multiple sub-tasks, perform task sensitivity detection on each sub-task, and then schedule models based on the sensitivity of each sub-task, thereby improving the effect of fine-grained control and scheduling.

[0120] Based on this, the operation and maintenance request is parsed to obtain request metadata, and task sensitivity detection is performed on the request metadata to determine the task type corresponding to the operation and maintenance request, which may include:

[0121] S51. Decompose the operation and maintenance request into tasks to obtain at least two sub-tasks that need to be executed to process the operation and maintenance request, and extract the task metadata of the sub-tasks.

[0122] In this embodiment, a pre-trained task decomposition model can be used to decompose maintenance requests into tasks. The task decomposition model is a pre-trained machine learning model capable of analyzing and decomposing maintenance requests into tasks. Since maintenance requests are typically in natural language form, the task decomposition model can be a language processing model, such as a large language model. Furthermore, if the maintenance request can be in other forms such as images, the task decomposition model can also be a multimodal processing model.

[0123] S52. Perform task sensitivity detection on task metadata to determine the task type corresponding to each subtask, so as to schedule the subtask to the local pre-trained model or the cloud pre-trained model for processing according to the task type.

[0124] Based on the above embodiments, after setting up a local storage pool and a cloud storage pool, this embodiment can further utilize the data in the local storage pool and the cloud storage pool to fine-tune the general model, thereby obtaining a local pre-trained model and a cloud pre-trained model. Based on this, the method may further include:

[0125] S61. Use the highly sensitive data in the local storage pool to set the first training data, and use the first training data to fine-tune the local pre-trained model.

[0126] It should be noted that this embodiment does not limit the setting method of the first training data or the specific method of fine-tuning the training. For example, if the general model is a language processing model, this embodiment can set instructions to fine-tune the training data based on highly sensitive data, and use the instructions to fine-tune the training data to fine-tune the general model. The following is one form of instruction-based fine-tuning of training data:

[0127] Input: Device status + log information + user query;

[0128] Output: Configuration suggestions / command snippets / troubleshooting steps.

[0129] S62. Use low-sensitivity data in the cloud storage pool to set up the second training data, use the second training data to fine-tune the cloud pre-trained model, add noise to the training gradient generated by the cloud pre-trained model during the training process, and update the parameters of the cloud pre-trained model based on the training gradient with added noise.

[0130] It should be noted that this embodiment does not limit the setting method of the second training data or the specific method of fine-tuning the training. For example, if the general model is a language processing model, this embodiment can set instructions to fine-tune the training data based on low-sensitivity data, and use the instructions to fine-tune the training data to fine-tune the general model.

[0131] It is worth noting that, to protect sensitive information during model training, this embodiment can add noise to the training gradients generated by the cloud-based pre-trained model during training, and update the parameters of the cloud-based pre-trained model based on the noise-added training gradients. For example, DP-SGD or similar algorithms can be used to inject noise to ensure privacy and security during the training process.

[0132] The following section will provide a complete explanation of the above maintenance and repair methods based on specific diagrams. Please refer to the provided text. Figure 3 , Figure 3 This is a schematic diagram of an operation and maintenance processing framework provided in an embodiment of the present invention. The operation and maintenance processing framework may include four parts: data source, fine-tuning training, user interaction, and automated execution closed loop.

[0133] In the data source section, the data collector can obtain service data, router logs, and telemetry data from network devices (such as APs) from third-party clouds, switches, and remote devices, and perform data cleaning and desensitization.

[0134] In the fine-tuning training part, a unified data lake / data warehouse can first be built using data collected by the data collector, and training samples can then be constructed. Subsequently, the training samples can be used to fine-tune the training of a general model (such as LLM) to obtain local and cloud models.

[0135] In terms of user interaction, the dialogue interface can receive operation and maintenance requests sent by users, perform sensitivity detection on the operation and maintenance requests, and schedule highly sensitive tasks to be processed by the local model and low-sensitivity tasks to be processed by the cloud model.

[0136] The automated execution closed-loop part can be implemented by the target system providing the execution script interface. When the local model generates the corresponding automated execution script, it can call the interface to execute the script, thereby realizing the automated operation and maintenance of the target system equipment (such as network equipment).

[0137] Please refer to Figure 4 , Figure 4 This is a schematic diagram illustrating local and cloud deployment according to an embodiment of the present invention. To enhance the protection of sensitive data, this embodiment can be deployed independently locally and in the cloud. Sensitive data (such as logs, configurations, etc.) can be stored in a local data lake. Data in the local data lake is used to train a local model, which handles highly sensitive issues. Non-sensitive data and aggregated data can be stored in a cloud data lake. Data in the cloud data lake is used to train a cloud model, which handles low-sensitivity issues. A dialog interface is used to route user maintenance requests. It can perform sensitivity detection on maintenance requests. If a maintenance request is determined to be a highly sensitive task, it can also perform risk detection on the user's request context. Only when the maintenance request is determined to correspond to a highly sensitive task with low risk will it be scheduled to be processed by the local model. If a maintenance request is determined to be a low-sensitivity task, the dialog interface can schedule the low-sensitivity task to be processed by the cloud model.

[0138] Please refer to Figure 5 , Figure 5 This is a timing diagram of an operation and maintenance processing method provided in an embodiment of the present invention, which fully illustrates the interaction process between the data collector, data lake, fine-tuning training module, local model, cloud model, dialogue interface, and user terminal.

[0139] The following describes the operation and maintenance processing device, electronic device, computer-readable storage medium, and computer program product provided in the embodiments of the present invention. The operation and maintenance processing device, electronic device, computer-readable storage medium, and computer program product described below can be referred to in correspondence with the operation and maintenance processing method described above.

[0140] Please refer to Figure 6 , Figure 6 This is a structural block diagram of an operation and maintenance processing device provided in an embodiment of the present invention. The device may include:

[0141] The classification storage module 601 is used to acquire system data of the target system, classify the system data into high-sensitivity data and low-sensitivity data, and store the high-sensitivity data in the local storage pool and the low-sensitivity data in the cloud storage pool.

[0142] The request parsing module 602 is used to parse the operation and maintenance request to obtain request metadata when it receives the operation and maintenance request sent by the user, and to perform task sensitivity detection on the request metadata to determine the task type corresponding to the operation and maintenance request.

[0143] The high-sensitivity task processing module 603 is used to generate operation and maintenance dialogue results and send them to the user terminal based on the operation and maintenance request and the high-sensitivity data in the local storage pool if the task type is a high-sensitivity task, or to perform operation and maintenance operations on the target system based on the operation and maintenance request and the high-sensitivity data using the local pre-trained model; wherein the target system, the local storage pool, and the local pre-trained model are all located in the same security domain.

[0144] The low-sensitivity task processing module 604 is used to generate operation and maintenance dialogue results and send them to the user terminal based on the operation and maintenance request and low-sensitivity data in the cloud storage pool if the task type is a low-sensitivity task.

[0145] Optionally, the request parsing module 602 includes:

[0146] The intent recognition submodule is used to recognize the intent of operation and maintenance requests and obtain user intent metadata.

[0147] The field parsing submodule is used to parse the fields of operation and maintenance requests, extract sensitive fields from the operation and maintenance requests, and obtain data-sensitive metadata.

[0148] The user information query submodule is used to query the user terminal information and request behavior information of the user terminal corresponding to the operation and maintenance request, and to use the user terminal information and request behavior information as network context metadata.

[0149] The integration submodule is used to integrate user intent metadata, data-sensitive metadata, and network context metadata into request metadata.

[0150] Optionally, the request parsing module 602 includes:

[0151] The intent matching submodule is used to perform semantic similarity matching between user intent metadata and preset intent information, determine the target intent information that best matches the user intent metadata, and use the matching degree of the target intent information as the intent matching degree.

[0152] The sensitive field ratio calculation submodule is used to calculate the ratio of the number of sensitive fields to the total number of fields in the operation and maintenance requests, and obtain the field sensitivity.

[0153] The behavior pattern matching submodule is used to perform behavior pattern matching with network context metadata and preset behavior information, determine the preset behavior information that best matches the network context metadata, and use the matching degree of the preset behavior information as the behavior pattern matching degree.

[0154] The weighted submodule is used to perform weighted fusion of intent matching degree, field sensitivity and behavior pattern matching degree using the preset weights corresponding to each of them, and obtain the sensitivity value.

[0155] The determination submodule is used to determine the task type as a low-sensitivity task if the sensitivity value is less than the first preset threshold, and to determine the task type as a high-sensitivity task if the sensitivity value is not less than the first preset threshold.

[0156] Optionally, the device may further include:

[0157] The adjustment module is used to adjust preset behavior information, sensitive fields to be extracted, preset behavior information, or preset weights when it receives input adjustment information.

[0158] Optionally, the device may further include:

[0159] The risk detection module is used to perform risk detection on the request metadata using a pre-trained risk detection model to determine the risk value corresponding to the operation and maintenance request. If the risk value is less than the second preset threshold, the module proceeds to the next step: generating operation and maintenance dialogue results based on the operation and maintenance request and highly sensitive data in the local storage pool using a local pre-trained model and sending them to the user terminal; or performing operation and maintenance operations on the target system based on the operation and maintenance request and highly sensitive data using a local pre-trained model. If the risk value is not less than the second preset threshold, the operation and maintenance request is intercepted.

[0160] Optionally, the device may further include:

[0161] The risk detection update module is used to generate training data using request metadata and manually labeled information; and to incrementally train the pre-trained risk detection model using the training data.

[0162] Optionally, the device may further include:

[0163] The task decomposition module is used to decompose the operation and maintenance request into tasks, obtain at least two sub-tasks that need to be executed to process the operation and maintenance request, and extract the task metadata of the sub-tasks.

[0164] The subtask sensitivity detection module is used to perform task sensitivity detection on task metadata, determine the task type corresponding to each subtask, and schedule the subtask to the local pre-trained model or the cloud pre-trained model for processing according to the task type.

[0165] Optionally, the classification storage module 601 may include:

[0166] The desensitization module is used to desensitize sensitive fields in low-sensitivity data;

[0167] The desensitized storage module is used to save the desensitized, low-sensitivity data to a cloud storage pool.

[0168] Optionally, the device may further include:

[0169] Local model fine-tuning training is used to set up the first training data using highly sensitive data in the local storage pool, and to fine-tune the local pre-trained model using the first training data.

[0170] The cloud-based model fine-tuning training is used to set up second training data using low-sensitivity data in the cloud storage pool, fine-tun the cloud-based pre-trained model using the second training data, add noise to the training gradient generated by the cloud-based pre-trained model during the training process, and update the parameters of the cloud-based pre-trained model based on the training gradient with added noise.

[0171] Please refer to Figure 7 , Figure 7 This is a structural block diagram of an electronic device provided in an embodiment of the present invention. The embodiment of the present invention provides an electronic device 70, including a processor 71 and a memory 72; wherein, the memory 72 is used to store a computer program; the processor 71 is used to execute the operation and maintenance processing method provided in the foregoing embodiment when executing the computer program.

[0172] For details regarding the specific procedures of the above-mentioned operation and maintenance methods, please refer to the relevant content provided in the foregoing embodiments, which will not be repeated here.

[0173] Furthermore, the memory 72, as a carrier for resource storage, can be a read-only memory, random access memory, disk, or optical disk, and the storage method can be temporary storage or permanent storage.

[0174] In addition, the electronic device 70 also includes a power supply 73, a communication interface 74, an input / output interface 75, and a communication bus 76; wherein, the power supply 73 is used to provide operating voltage for the various hardware devices on the electronic device 70; the communication interface 74 can create a data transmission channel between the electronic device 70 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this invention, and is not specifically limited here; the input / output interface 75 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0175] This invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the operation and maintenance processing method described in the above embodiments.

[0176] Since the embodiments of the computer program product section correspond to the embodiments of the operation and maintenance method section, please refer to the description of the embodiments of the operation and maintenance method section for the embodiments of the computer program product section, and they will not be repeated here.

[0177] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the operation and maintenance processing method described in the above embodiments.

[0178] Since the embodiments of the computer-readable storage medium portion correspond to the embodiments of the operation and maintenance method portion, the embodiments of the storage medium portion are described in the description of the embodiments of the operation and maintenance method portion, and will not be repeated here.

[0179] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0180] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0181] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0182] The above provides a detailed description of the operation and maintenance processing method, apparatus, electronic device, and storage medium provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of the present invention. It should be noted that those skilled in the art can make various improvements and modifications to the present invention without departing from its principles, and these improvements and modifications also fall within the protection scope of the present invention.

Claims

1. A method for operation and maintenance, characterized in that, include: The system data of the target system is acquired, classified into highly sensitive data and low sensitive data, and the highly sensitive data is stored in a local storage pool and the low sensitive data is stored in a cloud storage pool. When a maintenance request is received from a user, the maintenance request is parsed to obtain request metadata, and task sensitivity detection is performed on the request metadata to determine the task type corresponding to the maintenance request. If the task type is a highly sensitive task, the local pre-trained model is used to generate an operation and maintenance dialogue result based on the operation and maintenance request and the highly sensitive data in the local storage pool and send it to the user terminal, or the local pre-trained model is used to perform operation and maintenance operations on the target system based on the operation and maintenance request and the highly sensitive data; wherein, the target system, the local storage pool, and the local pre-trained model are all located in the same security domain; If the task type is a low-sensitivity task, the cloud-pre-trained model is used to generate an operation and maintenance dialogue result based on the operation and maintenance request and the low-sensitivity data in the cloud storage pool, and then the result is sent to the user terminal.

2. The operation and maintenance method according to claim 1, characterized in that, The parsing of the operation and maintenance request to obtain request metadata includes: The operation and maintenance request is subjected to intent recognition to obtain user intent metadata; The operation and maintenance request is parsed to extract sensitive fields and obtain sensitive metadata. Query the client information and request behavior information of the user terminal corresponding to the maintenance request, and use the client information and request behavior information as network context metadata; The user intent metadata, the data-sensitive metadata, and the network context metadata are integrated into the request metadata.

3. The operation and maintenance method according to claim 2, characterized in that, The step of performing task sensitivity detection on the request metadata to determine the task type corresponding to the operation and maintenance request includes: The semantic similarity of the user intent metadata with the preset intent information is used to determine the target intent information that best matches the user intent metadata, and the matching degree of the target intent information is used as the intent matching degree. The field sensitivity is obtained by calculating the ratio of the number of sensitive fields to the total number of fields in the maintenance request. The network context metadata is used to perform behavior pattern matching with preset behavior information to determine the preset behavior information that best matches the network context metadata, and the matching degree of the preset behavior information is used as the behavior pattern matching degree. Using the preset weights corresponding to the intent matching degree, the field sensitivity, and the behavior pattern matching degree, the intent matching degree, field sensitivity, and behavior pattern matching degree are weighted and fused to obtain a sensitivity value; If the sensitivity value is less than the first preset threshold, then the task type is determined to be a low-sensitivity task; If the sensitivity value is not less than the first preset threshold, then the task type is determined to be a highly sensitive task.

4. The operation and maintenance method according to claim 3, characterized in that, Also includes: When the input adjustment information is received, the preset behavior information, the sensitive field to be extracted, the preset behavior information, or the preset weight is adjusted using the adjustment information.

5. The operation and maintenance processing method according to claim 2, characterized in that, Before generating maintenance dialogue results based on the maintenance request and highly sensitive data in the local storage pool using a local pre-trained model and sending them to the user terminal, or before performing maintenance operations on the target system using the local pre-trained model based on the maintenance request and the highly sensitive data, the method further includes: A pre-trained risk detection model is used to detect risks in the request metadata to determine the risk value corresponding to the operation and maintenance request. If the risk value is less than the second preset threshold, then proceed to the step of generating an operation and maintenance dialogue result based on the operation and maintenance request and the highly sensitive data in the local storage pool using the local pre-trained model and sending it to the user terminal, or performing operation and maintenance operations on the target system based on the operation and maintenance request and the highly sensitive data using the local pre-trained model. If the risk value is not less than the second preset threshold, the maintenance request will be intercepted.

6. The operation and maintenance processing method according to claim 5, characterized in that, Also includes: Training data is generated using the requested metadata and manually labeled information; The pre-trained risk detection model is incrementally trained using the training data.

7. The operation and maintenance method according to claim 1, characterized in that, The process involves parsing the maintenance request to obtain request metadata, performing task sensitivity detection on the request metadata, and determining the task type corresponding to the maintenance request, including: The operation and maintenance request is decomposed into tasks to obtain at least two sub-tasks that need to be executed to process the operation and maintenance request, and the task metadata of the sub-tasks is extracted. The task metadata is subjected to task sensitivity detection to determine the task type corresponding to each subtask, so that the subtask is scheduled to be processed by the local pre-trained model or the cloud pre-trained model according to the task type.

8. The operation and maintenance method according to claim 1, characterized in that, The step of storing the low-sensitivity data in a cloud storage pool includes: Sensitive fields in the low-sensitivity data are de-identified; The desensitized, low-sensitivity data is saved to the cloud storage pool.

9. The operation and maintenance method according to claim 1, characterized in that, Also includes: The first training data is set using the highly sensitive data in the local storage pool, and the local pre-trained model is fine-tuned using the first training data. The second training data is set using low-sensitivity data in the cloud storage pool. The cloud pre-trained model is fine-tuned using the second training data. During the training process of the cloud pre-trained model, noise is added to the training gradient generated by the cloud pre-trained model, and the parameters of the cloud pre-trained model are updated based on the training gradient with added noise.

10. An operation and maintenance processing device, characterized in that, include: A classification storage module is used to acquire system data of the target system, classify the system data into high-sensitivity data and low-sensitivity data, and store the high-sensitivity data in a local storage pool and the low-sensitivity data in a cloud storage pool. The request parsing module is used to parse the operation and maintenance request sent by the user to obtain request metadata, and to perform task sensitivity detection on the request metadata to determine the task type corresponding to the operation and maintenance request. The high-sensitivity task processing module is used to generate an operation and maintenance dialogue result based on the operation and maintenance request and the high-sensitivity data in the local storage pool and send it to the user terminal if the task type is a high-sensitivity task, or to perform operation and maintenance operations on the target system based on the operation and maintenance request and the high-sensitivity data using the local pre-trained model. The low-sensitivity task processing module is used to generate an operation and maintenance dialogue result based on the operation and maintenance request and the low-sensitivity data in the cloud storage pool, and send it to the user terminal if the task type is a low-sensitivity task.

11. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the operation and maintenance processing method as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when loaded and executed by a processor, implement the operation and maintenance processing method as described in any one of claims 1 to 9.

Images